=================================================================== RCS file: /home/cvs/OpenXM_contrib2/asir2000/engine/nd.c,v retrieving revision 1.232 retrieving revision 1.241 diff -u -p -r1.232 -r1.241 --- OpenXM_contrib2/asir2000/engine/nd.c 2017/01/08 03:05:39 1.232 +++ OpenXM_contrib2/asir2000/engine/nd.c 2017/09/16 08:02:35 1.241 @@ -1,4 +1,4 @@ -/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.231 2016/12/05 10:29:14 noro Exp $ */ +/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.240 2017/09/15 01:52:51 noro Exp $ */ #include "nd.h" @@ -60,11 +60,12 @@ static int nd_module_rank,nd_poly_weight_len; static int *nd_poly_weight,*nd_module_weight; static NODE nd_tracelist; static NODE nd_alltracelist; -static int nd_gentrace,nd_gensyz,nd_nora,nd_newelim,nd_intersect; +static int nd_gentrace,nd_gensyz,nd_nora,nd_newelim,nd_intersect,nd_lf; static int *nd_gbblock; static NODE nd_nzlist,nd_check_splist; static int nd_splist; static int *nd_sugarweight; +static int nd_f4red,nd_rank0,nd_last_nonzero; NumberField get_numberfield(); UINT *nd_det_compute_bound(NDV **dm,int n,int j); @@ -97,9 +98,16 @@ void Pox_cmo_rpc(NODE,Obj *); ND nd_add_lf(ND p1,ND p2); void nd_mul_c_lf(ND p,GZ mul); void ndv_mul_c_lf(NDV p,GZ mul); -NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, +NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, + NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz); +NODE nd_f4_red_mod64_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, + NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz); +NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,int trace,UINT *s0vect,int col, NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred); int nd_gauss_elim_lf(mpz_t **mat0,int *sugar,int row,int col,int *colstat); +NODE nd_f4_lf_trace_main(int m,int **indp); +void nd_f4_lf_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp); + extern int lf_lazy; extern GZ current_mod_lf; @@ -648,6 +656,7 @@ int ndl_composite_compare(UINT *d1,UINT *d2) else if ( nd_work_vector[j] < 0 ) return -1; break; case 2: + end = start+len; for ( j = start; j < end; j++ ) if ( nd_work_vector[j] > 0 ) return 1; else if ( nd_work_vector[j] < 0 ) return -1; @@ -1452,7 +1461,7 @@ int nd_nf(int mod,ND d,ND g,NDV *ps,int full,NDC dn,ND sugar0 = sugar = SG(g); n = NV(g); - mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); + mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); if ( d ) for ( tail = BDY(d); NEXT(tail); tail = NEXT(tail) ); for ( ; g; ) { @@ -1540,7 +1549,7 @@ int nd_nf_pbucket(int mod,ND g,NDV *ps,int full,ND *rp bucket = create_pbucket(); add_pbucket(mod,bucket,g); d = 0; - mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); + mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); while ( 1 ) { if ( mod > 0 || mod == -1 ) hindex = head_pbucket(mod,bucket); @@ -1650,6 +1659,7 @@ again: r = ndv_dup_realloc((NDV)BDY(t),obpe,oadv,oepos); else r = (NDV)BDY(t); + if ( m ) ndv_mod(m,r); d = ndvtond(m,r); stat = nd_nf(m,0,d,nd_ps,0,0,&nf); if ( !stat ) { @@ -2662,7 +2672,7 @@ ND_pairs crit_B( ND_pairs d, int s ) t = DL(nd_psh[s]); prev = 0; head = cur = d; - lcm = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + lcm = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); while ( cur ) { tl = cur->lcm; if ( ndl_reducible(tl,t) ) { @@ -2903,6 +2913,18 @@ ND_pairs nd_minsugarp( ND_pairs d, ND_pairs *prest ) return dm0; } +int nd_tdeg(NDV c) +{ + int wmax = 0; + int i,len; + NMV a; + + len = LEN(c); + for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) ) + wmax = MAX(TD(DL(a)),wmax); + return wmax; +} + int ndv_newps(int m,NDV a,NDV aq,int f4) { int len; @@ -2927,19 +2949,35 @@ int ndv_newps(int m,NDV a,NDV aq,int f4) nd_ps[nd_psn] = a; if ( aq ) { nd_ps_trace[nd_psn] = aq; - if ( !nd_vc ) nd_ps_gz[nd_psn] = ndvtondvgz(aq); - register_hcf(aq); + if ( !m ) { + if ( !nd_vc ) nd_ps_gz[nd_psn] = ndvtondvgz(aq); + register_hcf(aq); + } else if ( m == -2 ) { + /* do nothing */ + } else + error("ndv_newps : invalud modulus"); nd_bound[nd_psn] = ndv_compute_bound(aq); - SG(r) = SG(aq); ndl_copy(HDL(aq),DL(r)); +#if 1 + SG(r) = SG(aq); +#else + SG(r) = nd_tdeg(aq); +#endif + ndl_copy(HDL(aq),DL(r)); } else { if ( !m ) register_hcf(a); nd_bound[nd_psn] = ndv_compute_bound(a); - SG(r) = SG(a); ndl_copy(HDL(a),DL(r)); +#if 1 + SG(r) = SG(a); +#else + SG(r) = nd_tdeg(a); +#endif + ndl_copy(HDL(a),DL(r)); if ( !m && !nd_vc ) nd_ps_gz[nd_psn] = ndvtondvgz(a); } if ( nd_demand ) { if ( aq ) { ndv_save(nd_ps_trace[nd_psn],nd_psn); + nd_ps_sym[nd_psn] = ndv_symbolic(m,nd_ps_trace[nd_psn]); nd_ps_trace_sym[nd_psn] = ndv_symbolic(m,nd_ps_trace[nd_psn]); nd_ps_trace[nd_psn] = 0; } else { @@ -2979,7 +3017,7 @@ int ndv_setup(int mod,int trace,NODE f,int dont_sort,i nd_tracelist = 0; for ( nd_psn = 0, s = f; s; s = NEXT(s) ) if ( BDY(s) ) nd_psn++; - w = (NDVI)ALLOCA(nd_psn*sizeof(struct oNDVI)); + w = (NDVI)MALLOC(nd_psn*sizeof(struct oNDVI)); for ( i = j = 0, s = f; s; s = NEXT(s), j++ ) if ( BDY(s) ) { w[i].p = BDY(s); w[i].i = j; i++; } if ( !dont_sort ) { @@ -3012,15 +3050,28 @@ int ndv_setup(int mod,int trace,NODE f,int dont_sort,i for ( i = 0; i < nd_psn; i++ ) { hc = HCU(w[i].p); if ( trace ) { - a = nd_ps_trace[i] = ndv_dup(0,w[i].p); - if ( !nd_vc ) nd_ps_gz[i] = ndvtondvgz(a); - if ( !dont_removecont) ndv_removecont(0,a); - register_hcf(a); - am = nd_ps[i] = ndv_dup(mod,a); - ndv_mod(mod,am); - if ( DL_COMPARE(HDL(am),HDL(a)) ) - return 0; - ndv_removecont(mod,am); + if ( mod == -2 ) { + /* over a large finite field */ + /* trace = small modulus */ + a = nd_ps_trace[i] = ndv_dup(-2,w[i].p); + ndv_mod(-2,a); + if ( !dont_removecont) ndv_removecont(-2,a); + am = nd_ps[i] = ndv_dup(trace,w[i].p); + ndv_mod(trace,am); + if ( DL_COMPARE(HDL(am),HDL(a)) ) + return 0; + ndv_removecont(trace,am); + } else { + a = nd_ps_trace[i] = ndv_dup(0,w[i].p); + if ( !nd_vc ) nd_ps_gz[i] = ndvtondvgz(a); + if ( !dont_removecont) ndv_removecont(0,a); + register_hcf(a); + am = nd_ps[i] = ndv_dup(mod,a); + ndv_mod(mod,am); + if ( DL_COMPARE(HDL(am),HDL(a)) ) + return 0; + ndv_removecont(mod,am); + } } else { a = nd_ps[i] = ndv_dup(mod,w[i].p); if ( !mod && !nd_vc ) nd_ps_gz[i] = ndvtondvgz(a); @@ -3039,6 +3090,7 @@ int ndv_setup(int mod,int trace,NODE f,int dont_sort,i if ( nd_demand ) { if ( trace ) { ndv_save(nd_ps_trace[i],i); + nd_ps_sym[i] = ndv_symbolic(mod,nd_ps_trace[i]); nd_ps_trace_sym[i] = ndv_symbolic(mod,nd_ps_trace[i]); nd_ps_trace[i] = 0; } else { @@ -3094,10 +3146,10 @@ void preprocess_algcoef(VL vv,VL av,struct order_spec if ( NID(hc) == N_DA ) { invdalg(hc,&inv); for ( m = BDY(d); m; m = NEXT(m) ) { - muldalg(inv,(DAlg)m->c,&da); m->c = (P)da; + muldalg(inv,(DAlg)m->c,&da); m->c = (Obj)da; } } - initd(ord); dtop(vv,vv,d,&poly); BDY(f) = (pointer)poly; + initd(ord); dtop(vv,vv,d,(Obj *)&poly); BDY(f) = (pointer)poly; } obj_dalgtoalg((Obj)f1,(Obj *)&f); @@ -3176,7 +3228,7 @@ void nd_gr(LIST f,LIST v,int m,int homo,int retdp,int NODE tr,tl1,tl2,tl3,tl4,nzlist; LIST l1,l2,l3,l4,l5; int j; - Q jq,bpe; + Q jq,bpe,last_nonzero; int *perm; EPOS oepos; int obpe,oadv,ompos,cbpe; @@ -3286,7 +3338,7 @@ void nd_gr(LIST f,LIST v,int m,int homo,int retdp,int } return; } - x = f4?nd_f4(m,&perm):nd_gb(m,ishomo || homo,0,0,&perm); + x = f4?nd_f4(m,0,&perm):nd_gb(m,ishomo || homo,0,0,&perm); if ( !x ) { *rp = 0; return; } @@ -3335,7 +3387,9 @@ FINAL: if ( nd_gentrace ) { if ( f4 ) { STOQ(16,bpe); - tr = mknode(4,*rp,(!ishomo&&homo)?ONE:0,BDY(nzlist),bpe); MKLIST(*rp,tr); + STOQ(nd_last_nonzero,last_nonzero); + tr = mknode(5,*rp,(!ishomo&&homo)?ONE:0,BDY(nzlist),bpe,last_nonzero); MKLIST(*rp,tr); + } else { tl1 = reverse_node(tl1); tl2 = reverse_node(tl2); tl3 = reverse_node(tl3); @@ -3466,7 +3520,7 @@ NDV recompute_trace(NODE ti,NDV *p,int mod) static int afo=0; afo++; - mul = (NM)ALLOCA(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); + mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); CM(mul) = 1; tail = 0; for ( i = 0, d = r = 0; ti; ti = NEXT(ti), i++ ) { @@ -3626,7 +3680,15 @@ void nd_gr_trace(LIST f,LIST v,int trace,int homo,int Q jq,bpe; nd_module = 0; + nd_lf = 0; parse_nd_option(current_option); + if ( nd_lf ) { + if ( f4 ) + nd_f4_lf_trace(f,v,trace,homo,ord,rp); + else + error("nd_gr_trace is not implemented yet over a large finite field"); + return; + } if ( DP_Multiple ) nd_scale = ((double)DP_Multiple)/(double)(Denominator?Denominator:1); @@ -3753,7 +3815,10 @@ void nd_gr_trace(LIST f,LIST v,int trace,int homo,int tl3 = nd_alltracelist; nd_alltracelist = 0; } else tl3 = 0; /* gbcheck : cand is a GB of Id(cand) ? */ - ret = nd_gb(0,0,1,nd_gensyz?1:0,0)!=0; + if ( nd_vc || nd_gentrace || nd_gensyz ) + ret = nd_gb(0,0,1,nd_gensyz?1:0,0)!=0; + else + ret = nd_f4(0,1,0)!=0; if ( nd_gentrace && nd_gensyz ) { tl4 = nd_alltracelist; nd_alltracelist = 0; } else tl4 = 0; @@ -3777,7 +3842,7 @@ void nd_gr_trace(LIST f,LIST v,int trace,int homo,int } get_eg(&eg1); init_eg(&eg_check); add_eg(&eg_check,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"check=%fsec\n",eg_check.exectime+eg_check.gctime); + fprintf(asir_out,"check=%.3fsec,",eg_check.exectime+eg_check.gctime); /* dp->p */ nd_bpe = cbpe; nd_setup_parameters(nd_nvar,0); @@ -3866,7 +3931,7 @@ void nmtodp(int mod,NM m,DP *r) NEWMP(mr); mr->dl = ndltodl(nd_nvar,DL(m)); - mr->c = ndctop(mod,m->c); + mr->c = (Obj)ndctop(mod,m->c); NEXT(mr) = 0; MKDP(nd_nvar,mr,dp); dp->sugar = mr->dl->td; *r = dp; } @@ -3891,7 +3956,7 @@ void ndl_print(UINT *dl) for ( i = 0; i < n; i++ ) printf(i==n-1?"%d":"%d,",GET_EXP(dl,i)); } printf(">>"); - if ( MPOS(dl) ) + if ( nd_module && MPOS(dl) ) printf("*e%d",MPOS(dl)); } @@ -3953,7 +4018,7 @@ void nd_removecont(int mod,ND p) } else if ( mod ) nd_mul_c(mod,p,invm(HCM(p),mod)); else { for ( m = BDY(p), n = 0; m; m = NEXT(m), n++ ); - w = (Q *)ALLOCA(n*sizeof(Q)); + w = (Q *)MALLOC(n*sizeof(Q)); v.len = n; v.body = (pointer *)w; for ( m = BDY(p), i = 0; i < n; m = NEXT(m), i++ ) w[i] = CQ(m); @@ -3974,7 +4039,7 @@ void nd_removecont2(ND p1,ND p2) n1 = nd_length(p1); n2 = nd_length(p2); n = n1+n2; - w = (Q *)ALLOCA(n*sizeof(Q)); + w = (Q *)MALLOC(n*sizeof(Q)); v.len = n; v.body = (pointer *)w; i = 0; @@ -4009,8 +4074,8 @@ void ndv_removecont(int mod,NDV p) ndv_mul_c(mod,p,invm(HCM(p),mod)); else { len = p->len; - w = (P *)ALLOCA(len*sizeof(P)); - c = (Q *)ALLOCA(len*sizeof(Q)); + w = (P *)MALLOC(len*sizeof(P)); + c = (Q *)MALLOC(len*sizeof(Q)); for ( m = BDY(p), all_p = 1, i = 0; i < len; NMV_ADV(m), i++ ) { ptozp(CP(m),1,&c[i],&w[i]); all_p = all_p && !NUM(w[i]); @@ -4044,7 +4109,7 @@ void ndv_homogenize(NDV p,int obpe,int oadv,EPOS oepos mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p); m = (NMV)((char *)mr0+(len-1)*oadv); mr = (NMV)((char *)mr0+(len-1)*nmv_adv); - t = (NMV)ALLOCA(nmv_adv); + t = (NMV)MALLOC(nmv_adv); for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) { ndl_homogenize(DL(m),DL(t),obpe,oepos,ompos,max); CQ(mr) = CQ(m); @@ -4092,8 +4157,8 @@ void nd_heu_nezgcdnpz(VL vl,P *pl,int m,int full,P *pr NODE l; P h,gcd,t; - tpl = (P *)ALLOCA(m*sizeof(P)); - tpl1 = (P *)ALLOCA(m*sizeof(P)); + tpl = (P *)MALLOC(m*sizeof(P)); + tpl1 = (P *)MALLOC(m*sizeof(P)); bcopy(pl,tpl,m*sizeof(P)); gcd = (P)ONE; for ( l = nd_hcf; l; l = NEXT(l) ) { @@ -4129,8 +4194,8 @@ void removecont_array(P *p,int n,int full) if ( all_q ) { removecont_array_q((Q *)p,n); } else { - c = (Q *)ALLOCA(n*sizeof(Q)); - w = (P *)ALLOCA(n*sizeof(P)); + c = (Q *)MALLOC(n*sizeof(Q)); + w = (P *)MALLOC(n*sizeof(P)); for ( i = 0; i < n; i++ ) { ptozp(p[i],1,&c[i],&w[i]); } @@ -4150,8 +4215,8 @@ void removecont_array_q(Q *c,int n) N qn,rn,gn; Q *q,*r; - q = (Q *)ALLOCA(n*sizeof(Q)); - r = (Q *)ALLOCA(n*sizeof(Q)); + q = (Q *)MALLOC(n*sizeof(Q)); + r = (Q *)MALLOC(n*sizeof(Q)); v.id = O_VECT; v.len = n; v.body = (pointer *)c; igcdv_estimate(&v,&d0); for ( i = 0; i < n; i++ ) { @@ -4269,8 +4334,8 @@ UINT *ndv_compute_bound(NDV p) if ( !p ) return 0; - d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); - d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + d1 = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); + d2 = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); len = LEN(p); m = BDY(p); ndl_copy(DL(m),d1); NMV_ADV(m); for ( i = 1; i < len; i++, NMV_ADV(m) ) { @@ -4298,8 +4363,8 @@ UINT *nd_compute_bound(ND p) if ( !p ) return 0; - d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); - d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + d1 = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); + d2 = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); len = LEN(p); m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m); for ( m = NEXT(m); m; m = NEXT(m) ) { @@ -4332,7 +4397,11 @@ int nd_get_exporigin(struct order_spec *ord) /* module d[0]:weight d[1]:w0,...,d[nd_exporigin-2]:w(n-1) */ return ord->ord.block.length+1+nd_module; case 3: case 259: +#if 0 error("nd_get_exporigin : composite order is not supported yet."); +#else + return 1+nd_module; +#endif } } @@ -4458,7 +4527,7 @@ ND_pairs nd_reconstruct(int trace,ND_pairs d) ndl_reconstruct(LCM(t),LCM(s),obpe,oepos); } - old_red = (RHist *)ALLOCA(REDTAB_LEN*sizeof(RHist)); + old_red = (RHist *)MALLOC(REDTAB_LEN*sizeof(RHist)); for ( i = 0; i < REDTAB_LEN; i++ ) { old_red[i] = nd_red[i]; nd_red[i] = 0; @@ -4655,8 +4724,8 @@ ND weyl_ndv_mul_nm(int mod,NM m0,NDV p) { d0 = DL(m0); l = LEN(p); for ( i = 0, tlen = 1; i < n2; i++ ) tlen *= (GET_EXP(d0,n2+i)+1); - tab = (NM *)ALLOCA(tlen*sizeof(NM)); - psum = (NM *)ALLOCA(tlen*sizeof(NM)); + tab = (NM *)MALLOC(tlen*sizeof(NM)); + psum = (NM *)MALLOC(tlen*sizeof(NM)); for ( i = 0; i < tlen; i++ ) psum[i] = 0; m1 = (NMV)(((char *)BDY(p))+nmv_adv*(l-1)); for ( i = l-1; i >= 0; i--, NMV_PREV(m1) ) { @@ -4734,10 +4803,10 @@ void weyl_mul_nm_nmv(int n,int mod,NM m0,NMV m1,NM *ta } min = MIN(k,l); if ( mod ) { - ctab = (UINT *)ALLOCA((min+1)*sizeof(UINT)); + ctab = (UINT *)MALLOC((min+1)*sizeof(UINT)); mkwcm(k,l,mod,ctab); } else { - ctab_q = (Q *)ALLOCA((min+1)*sizeof(Q)); + ctab_q = (Q *)MALLOC((min+1)*sizeof(Q)); mkwc(k,l,ctab_q); } for ( j = min; j >= 0; j-- ) { @@ -4897,7 +4966,7 @@ ND nd_quo(int mod,PGeoBucket bucket,NDV d) else { nv = NV(d); mq0 = 0; - tm = (NMV)ALLOCA(nmv_adv); + tm = (NMV)MALLOC(nmv_adv); while ( 1 ) { if ( mod > 0 || mod == -1 ) hindex = head_pbucket(mod,bucket); @@ -4947,7 +5016,7 @@ void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos) mr0 = nmv_adv>oadv?(NMV)REALLOC(BDY(p),len*nmv_adv):BDY(p); m = (NMV)((char *)mr0+(len-1)*oadv); mr = (NMV)((char *)mr0+(len-1)*nmv_adv); - t = (NMV)ALLOCA(nmv_adv); + t = (NMV)MALLOC(nmv_adv); for ( i = 0; i < len; i++, NMV_OPREV(m), NMV_PREV(mr) ) { CQ(t) = CQ(m); for ( k = 0; k < nd_wpd; k++ ) DL(t)[k] = 0; @@ -5095,18 +5164,23 @@ void ndv_mod(int mod,NDV p) if ( mod == -1 ) for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) { simp_ff((Obj)CP(t),&gfs); - r = FTOIF(CONT((GFS)gfs)); - CM(d) = r; - ndl_copy(DL(t),DL(d)); - NMV_ADV(d); - dlen++; + if ( gfs ) { + r = FTOIF(CONT((GFS)gfs)); + CM(d) = r; + ndl_copy(DL(t),DL(d)); + NMV_ADV(d); + dlen++; + } } else if ( mod == -2 ) for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) { - simp_ff((Obj)CP(t),&gfs); lmtolf(gfs,&CZ(d)); - ndl_copy(DL(t),DL(d)); - NMV_ADV(d); - dlen++; + simp_ff((Obj)CP(t),&gfs); + if ( gfs ) { + lmtolf(gfs,&CZ(d)); + ndl_copy(DL(t),DL(d)); + NMV_ADV(d); + dlen++; + } } else for ( t = d = BDY(p), i = 0; i < len; i++, NMV_ADV(t) ) { @@ -5155,8 +5229,8 @@ void pltozpl(LIST l,Q *cont,LIST *pp) LIST r; nd = BDY(l); n = length(nd); - pl = (P *)ALLOCA(n*sizeof(P)); - cl = (Q *)ALLOCA(n*sizeof(P)); + pl = (P *)MALLOC(n*sizeof(P)); + cl = (Q *)MALLOC(n*sizeof(P)); for ( i = 0; i < n; i++, nd = NEXT(nd) ) ptozp((P)BDY(nd),1,&cl[i],&dmy); qltozl(cl,n,&dvr); @@ -5221,7 +5295,7 @@ ND ptond(VL vl,VL dvl,P p) return r; } else { for ( dc = DC(p), k = 0; dc; dc = NEXT(dc), k++ ); - w = (DCP *)ALLOCA(k*sizeof(DCP)); + w = (DCP *)MALLOC(k*sizeof(DCP)); for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc; for ( i = 0, tvl = dvl, v = VR(p); tvl && tvl->v != v; tvl = NEXT(tvl), i++ ); @@ -5271,7 +5345,7 @@ P ndvtop(int mod,VL vl,VL dvl,NDV p) if ( mod == -1 ) { e = IFTOF(CM(m)); MKGFS(e,gfs); c = (P)gfs; } else if ( mod == -2 ) { - c = gztoz(CZ(m)); + c = (P)gztoz(CZ(m)); } else if ( mod > 0 ) { STOQ(CM(m),q); c = (P)q; } else @@ -5303,7 +5377,7 @@ LIST ndvtopl(int mod,VL vl,VL dvl,NDV p,int rank) if ( !p ) return 0; else { - a = (P *)ALLOCA((rank+1)*sizeof(P)); + a = (P *)MALLOC((rank+1)*sizeof(P)); for ( i = 0; i <= rank; i++ ) a[i] = 0; len = LEN(p); n = NV(p); @@ -5390,7 +5464,7 @@ DP ndvtodp(int mod,NDV p) for ( t = BDY(p), i = 0; i < len; NMV_ADV(t), i++ ) { NEXTMP(m0,m); m->dl = ndltodl(nd_nvar,DL(t)); - m->c = ndctop(mod,t->c); + m->c = (Obj)ndctop(mod,t->c); } NEXT(m) = 0; MKDP(nd_nvar,m0,d); @@ -5411,7 +5485,7 @@ DP ndtodp(int mod,ND p) for ( t = BDY(p); t; t = NEXT(t) ) { NEXTMP(m0,m); m->dl = ndltodl(nd_nvar,DL(t)); - m->c = ndctop(mod,t->c); + m->c = (Obj)ndctop(mod,t->c); } NEXT(m) = 0; MKDP(nd_nvar,m0,d); @@ -5461,7 +5535,7 @@ NODE ndv_reducebase(NODE x,int *perm) NODE t,t0; len = length(x); - w = (NDVI)ALLOCA(len*sizeof(struct oNDVI)); + w = (NDVI)MALLOC(len*sizeof(struct oNDVI)); for ( i = 0, t = x; i < len; i++, t = NEXT(t) ) { w[i].p = BDY(t); w[i].i = perm[i]; } @@ -5750,6 +5824,27 @@ int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r) return i; } +#if defined(__GNUC__) + +#define MOD128(a,c,m) ((a)=(((c)!=0||((a)>=(m)))?(((((U128)(c))<<64)+(a))%(m)):(a))) + +int nd_to_vect64(int mod,UINT *s0,int n,ND d,U64 *r) +{ + NM m; + UINT *t,*s; + int i; + + for ( i = 0; i < n; i++ ) r[i] = 0; + for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) { + t = DL(m); + for ( ; !ndl_equal(t,s); s += nd_wpd, i++ ); + r[i] = (U64)CM(m); + } + for ( i = 0; !r[i]; i++ ); + return i; +} +#endif + int nd_to_vect_q(UINT *s0,int n,ND d,Q *r) { NM m; @@ -5809,7 +5904,7 @@ int nd_nm_to_vect_2(UINT *s0,int n,int *s0hash,NDV p,N d = DL(m); len = LEN(p); - t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + t = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) { ndl_add(d,DL(mr),t); h = ndl_hash_value(t); @@ -5834,7 +5929,7 @@ Q *nm_ind_pair_to_vect(int mod,UINT *s0,int n,NM_ind_p p = nd_ps[pair->index]; len = LEN(p); r = (Q *)CALLOC(n,sizeof(Q)); - t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); + t = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) { ndl_add(d,DL(mr),t); for ( ; !ndl_equal(t,s); s += nd_wpd, i++ ); @@ -5843,7 +5938,7 @@ Q *nm_ind_pair_to_vect(int mod,UINT *s0,int n,NM_ind_p return r; } -IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0,int n,int *s0hash,NM_ind_pair pair) +IndArray nm_ind_pair_to_vect_compress(int trace,UINT *s0,int n,int *s0hash,NM_ind_pair pair) { NM m; NMV mr; @@ -5858,10 +5953,14 @@ struct oEGT eg0,eg1; m = pair->mul; d = DL(m); - p = nd_demand?nd_ps_sym[pair->index]:nd_ps[pair->index]; + if ( trace ) + p = nd_demand?nd_ps_trace_sym[pair->index]:nd_ps_trace[pair->index]; + else + p = nd_demand?nd_ps_sym[pair->index]:nd_ps[pair->index]; + len = LEN(p); - t = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); - v = (unsigned int *)ALLOCA(len*sizeof(unsigned int)); + t = (UINT *)MALLOC(nd_wpd*sizeof(UINT)); + v = (unsigned int *)MALLOC(len*sizeof(unsigned int)); get_eg(&eg0); for ( i = j = 0, s = s0, mr = BDY(p); j < len; j++, NMV_ADV(mr) ) { ndl_add(d,DL(mr),t); @@ -5931,7 +6030,7 @@ int ndv_reduce_vect_q(Q *svect,int trace,int col,IndAr for ( i = 0; i < col && !svect[i]; i++ ); if ( i == col ) return maxrs; hmag = p_mag((P)svect[i])*nd_scale; - cvect = (Q *)ALLOCA(col*sizeof(Q)); + cvect = (Q *)MALLOC(col*sizeof(Q)); for ( i = 0; i < nred; i++ ) { ivect = imat[i]; k = ivect->head; @@ -6071,7 +6170,6 @@ int ndv_reduce_vect_gz(GZ *gvect,int trace,int col,Ind return maxrs; } - int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) { int i,j,k,len,pos,prev; @@ -6141,6 +6239,78 @@ int ndv_reduce_vect(int m,UINT *svect,int col,IndArray return maxrs; } +#if defined(__GNUC__) + +int ndv_reduce_vect64(int m,U64 *svect,U64 *cvect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) +{ + int i,j,k,len,pos,prev; + U64 a,c,c1,c2; + IndArray ivect; + unsigned char *ivc; + unsigned short *ivs; + unsigned int *ivi; + NDV redv; + NMV mr; + NODE rp; + int maxrs; + + for ( i = 0; i < col; i++ ) cvect[i] = 0; + maxrs = 0; + for ( i = 0; i < nred; i++ ) { + ivect = imat[i]; + k = ivect->head; + a = svect[k]; c = cvect[k]; + MOD128(a,c,m); + svect[k] = a; cvect[k] = 0; + if ( c = svect[k] ) { + maxrs = MAX(maxrs,rp0[i]->sugar); + c = m-c; redv = nd_ps[rp0[i]->index]; + len = LEN(redv); mr = BDY(redv); + svect[k] = 0; prev = k; + switch ( ivect->width ) { + case 1: + ivc = ivect->index.c; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivc[j]; c1 = CM(mr); prev = pos; + if ( c1 ) { + c2 = svect[pos]+c1*c; + if ( c2 < svect[pos] ) cvect[pos]++; + svect[pos] = c2; + } + } + break; + case 2: + ivs = ivect->index.s; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivs[j]; c1 = CM(mr); prev = pos; + if ( c1 ) { + c2 = svect[pos]+c1*c; + if ( c2 < svect[pos] ) cvect[pos]++; + svect[pos] = c2; + } + } + break; + case 4: + ivi = ivect->index.i; + for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { + pos = prev+ivi[j]; c1 = CM(mr); prev = pos; + if ( c1 ) { + c2 = svect[pos]+c1*c; + if ( c2 < svect[pos] ) cvect[pos]++; + svect[pos] = c2; + } + } + break; + } + } + } + for ( i = 0; i < col; i++ ) { + a = svect[i]; c = cvect[i]; MOD128(a,c,m); svect[i] = a; + } + return maxrs; +} +#endif + int ndv_reduce_vect_sf(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) { int i,j,k,len,pos,prev; @@ -6238,7 +6408,7 @@ ND nd_add_lf(ND p1,ND p2) } } -int ndv_reduce_vect_lf(mpz_t *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) +int ndv_reduce_vect_lf(mpz_t *svect,int trace,int col,IndArray *imat,NM_ind_pair *rp0,int nred) { int i,j,k,len,pos,prev; mpz_t c,mc,c1; @@ -6260,7 +6430,7 @@ int ndv_reduce_vect_lf(mpz_t *svect,int col,IndArray * if ( mpz_sgn(svect[k]) ) { maxrs = MAX(maxrs,rp0[i]->sugar); mpz_neg(svect[k],svect[k]); - redv = nd_ps[rp0[i]->index]; + redv = trace?nd_ps_trace[rp0[i]->index]:nd_ps[rp0[i]->index]; len = LEN(redv); mr = BDY(redv); prev = k; switch ( ivect->width ) { @@ -6303,10 +6473,12 @@ int nd_gauss_elim_lf(mpz_t **mat0,int *sugar,int row,i mpz_t *t,*pivot,*pk; mpz_t **mat; struct oEGT eg0,eg1,eg_forward,eg_mod,eg_back; + int size,size1; mpz_init(inv); mpz_init(a); mat = (mpz_t **)mat0; + size = 0; for ( rank = 0, j = 0; j < col; j++ ) { for ( i = rank; i < row; i++ ) { mpz_mod(mat[i][j],mat[i][j],BDY(current_mod_lf)); @@ -6396,6 +6568,36 @@ NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhea } } +#if defined(__GNUC__) +NDV vect64_to_ndv(U64 *vect,int spcol,int col,int *rhead,UINT *s0vect) +{ + int j,k,len; + UINT *p; + UINT c; + NDV r; + NMV mr0,mr; + + for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++; + if ( !len ) return 0; + else { + mr0 = (NMV)MALLOC_ATOMIC_IGNORE_OFF_PAGE(nmv_adv*len); +#if 0 + ndv_alloc += nmv_adv*len; +#endif + mr = mr0; + p = s0vect; + for ( j = k = 0; j < col; j++, p += nd_wpd ) + if ( !rhead[j] ) { + if ( c = (UINT)vect[k++] ) { + ndl_copy(p,DL(mr)); CM(mr) = c; NMV_ADV(mr); + } + } + MKNDV(nd_nvar,mr0,len,r); + return r; + } +} +#endif + NDV vect_to_ndv_2(unsigned long *vect,int col,UINT *s0vect) { int j,k,len; @@ -6564,6 +6766,7 @@ int nd_symbolic_preproc(PGeoBucket bucket,int trace,UI NM_ind_pair pair; ND red; NDV *ps; + static int afo; s0 = 0; rp0 = 0; col = 0; if ( nd_demand ) @@ -6581,7 +6784,8 @@ int nd_symbolic_preproc(PGeoBucket bucket,int trace,UI h = nd_psh[index]; NEWNM(mul); ndl_sub(DL(head),DL(h),DL(mul)); - if ( ndl_check_bound2(index,DL(mul)) ) return 0; + if ( ndl_check_bound2(index,DL(mul)) ) + return 0; sugar = TD(DL(mul))+SG(ps[index]); MKNM_ind_pair(pair,mul,index,sugar); red = ndv_mul_nm_symbolic(mul,ps[index]); @@ -6600,9 +6804,9 @@ int nd_symbolic_preproc(PGeoBucket bucket,int trace,UI return col; } -NODE nd_f4(int m,int **indp) +NODE nd_f4(int m,int checkonly,int **indp) { - int i,nh,stat,index; + int i,nh,stat,index,f4red; NODE r,g,tn0,tn,node; ND_pairs d,l,t,ll0,ll; LIST l0,l1; @@ -6632,11 +6836,13 @@ NODE nd_f4(int m,int **indp) } nzlist = 0; nzlist_t = nd_nzlist; + f4red = 1; + nd_last_nonzero = 0; while ( d || nzlist_t ) { get_eg(&eg0); if ( nd_nzlist ) { node = BDY((LIST)BDY(nzlist_t)); - sugar = (int)ARG0(node); + sugar = QTOS((Q)ARG0(node)); tn = BDY((LIST)ARG1(node)); if ( !tn ) { nzlist_t = NEXT(nzlist_t); @@ -6669,10 +6875,12 @@ NODE nd_f4(int m,int **indp) } get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"sugar=%d,symb=%fsec,", + fprintf(asir_out,"sugar=%d,symb=%.3fsec,", sugar,eg_f4.exectime+eg_f4.gctime); nflist = nd_f4_red(m,l,0,s0vect,col,rp0,nd_gentrace?&ll:0); + if ( checkonly && nflist ) return 0; /* adding new bases */ + if ( nflist ) nd_last_nonzero = f4red; for ( r = nflist; r; r = NEXT(r) ) { nf = (NDV)BDY(r); ndv_removecont(m,nf); @@ -6688,6 +6896,9 @@ NODE nd_f4(int m,int **indp) if ( !nd_nzlist ) d = update_pairs(d,g,nh,0); g = update_base(g,nh); } + if ( DP_Print ) { + fprintf(asir_out,"f4red=%d,gblen=%d\n",f4red,length(g)); fflush(asir_out); + } if ( nd_gentrace ) { for ( t = ll, tn0 = 0; t; t = NEXT(t) ) { NEXTNODE(tn0,tn); @@ -6700,6 +6911,9 @@ NODE nd_f4(int m,int **indp) MKNODE(node,l1,nzlist); nzlist = node; } if ( nd_nzlist ) nzlist_t = NEXT(nzlist_t); + f4red++; + if ( nd_f4red && f4red > nd_f4red ) break; + if ( nd_rank0 && !nflist ) break; } if ( nd_gentrace ) { MKLIST(l0,reverse_node(nzlist)); @@ -6761,7 +6975,7 @@ NODE nd_f4_trace(int m,int **indp) } get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); if ( DP_Print ) - fprintf(asir_out,"sugar=%d,symb=%fsec,", + fprintf(asir_out,"sugar=%d,symb=%.3fsec,", sugar,eg_f4.exectime+eg_f4.gctime); nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0); if ( !l0 ) continue; @@ -6813,109 +7027,6 @@ NODE nd_f4_trace(int m,int **indp) return g; } -NODE nd_f4_pseudo_trace(int m,int **indp) -{ - int i,nh,stat,index; - NODE r,g; - ND_pairs d,l,l0,t; - ND spol,red; - NDV nf,redv,nfqv,nfv; - NM s0,s; - NODE rp0,srp0,nflist; - int nsp,nred,col,rank,len,k,j,a; - UINT c; - UINT **spmat; - UINT *s0vect,*svect,*p,*v; - int *colstat; - IndArray *imat; - int *rhead; - int spcol,sprow; - int sugar; - PGeoBucket bucket; - struct oEGT eg0,eg1,eg_f4; - - g = 0; d = 0; - for ( i = 0; i < nd_psn; i++ ) { - d = update_pairs(d,g,i,0); - g = update_base(g,i); - } - while ( d ) { - get_eg(&eg0); - l = nd_minsugarp(d,&d); - sugar = SG(l); - bucket = create_pbucket(); - stat = nd_sp_f4(m,0,l,bucket); - if ( !stat ) { - for ( t = l; NEXT(t); t = NEXT(t) ); - NEXT(t) = d; d = l; - d = nd_reconstruct(1,d); - continue; - } - if ( bucket->m < 0 ) continue; - col = nd_symbolic_preproc(bucket,0,&s0vect,&rp0); - if ( !col ) { - for ( t = l; NEXT(t); t = NEXT(t) ); - NEXT(t) = d; d = l; - d = nd_reconstruct(1,d); - continue; - } - get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); - if ( DP_Print ) - fprintf(asir_out,"sugar=%d,symb=%fsec,", - sugar,eg_f4.exectime+eg_f4.gctime); - nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0); - if ( !l0 ) continue; - l = l0; - - /* over Q */ - while ( 1 ) { - bucket = create_pbucket(); - stat = nd_sp_f4(0,1,l,bucket); - if ( !stat ) { - for ( t = l; NEXT(t); t = NEXT(t) ); - NEXT(t) = d; d = l; - d = nd_reconstruct(1,d); - continue; - } - if ( bucket->m < 0 ) continue; - col = nd_symbolic_preproc(bucket,1,&s0vect,&rp0); - if ( !col ) { - for ( t = l; NEXT(t); t = NEXT(t) ); - NEXT(t) = d; d = l; - d = nd_reconstruct(1,d); - continue; - } - nflist = nd_f4_red(0,l,1,s0vect,col,rp0,0); - } - - /* adding new bases */ - for ( r = nflist; r; r = NEXT(r) ) { - nfqv = (NDV)BDY(r); - ndv_removecont(0,nfqv); - if ( !rem(NM(HCQ(nfqv)),m) ) return 0; - if ( nd_nalg ) { - ND nf1; - - nf1 = ndvtond(m,nfqv); - nd_monic(0,&nf1); - nd_removecont(0,nf1); - nfqv = ndtondv(0,nf1); nd_free(nf1); - } - nfv = ndv_dup(0,nfqv); - ndv_mod(m,nfv); - ndv_removecont(m,nfv); - nh = ndv_newps(0,nfv,nfqv,1); - d = update_pairs(d,g,nh,0); - g = update_base(g,nh); - } - } -#if 0 - fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); -#endif - conv_ilist(0,1,g,indp); - return g; -} - int rref(matrix mat,int *sugar) { int row,col,i,j,k,l,s,wcol,wj; @@ -6998,7 +7109,7 @@ void red_by_vect_2(matrix mat,int *sugar,unsigned long row = mat->row; col = mat->col; wcol = (col+BLEN-1)/BLEN; - pos = (int *)ALLOCA(wcol*sizeof(int)); + pos = (int *)MALLOC(wcol*sizeof(int)); bzero(pos,wcol*sizeof(int)); for ( i = j = 0; i < wcol; i++ ) if ( v[i] ) pos[j++] = i;; @@ -7038,11 +7149,11 @@ init_eg(&eg_search); for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); nred = length(rp0); mat = alloc_matrix(nsp,col); - s0hash = (int *)ALLOCA(col*sizeof(int)); + s0hash = (int *)MALLOC(col*sizeof(int)); for ( i = 0, s = s0vect; i < col; i++, s += nd_wpd ) s0hash[i] = ndl_hash_value(s); - sugar = (int *)ALLOCA(nsp*sizeof(int)); + sugar = (int *)MALLOC(nsp*sizeof(int)); for ( i = 0, sp = sp0; sp; sp = NEXT(sp) ) { nd_sp(2,0,sp,&spol); if ( spol ) { @@ -7078,7 +7189,7 @@ init_eg(&eg_search); get_eg(&eg2); init_eg(&eg_elim2); add_eg(&eg_elim2,&eg1,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,elim2=%fsec\n", + fprintf(asir_out,"elim1=%.3fsec,elim2=%.3fsec,", eg_elim1.exectime+eg_elim1.gctime,eg_elim2.exectime+eg_elim2.gctime); fflush(asir_out); } @@ -7103,34 +7214,47 @@ NODE nd_f4_red(int m,ND_pairs sp0,int trace,UINT *s0ve init_eg(&eg_search); for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); nred = length(rp0); - imat = (IndArray *)ALLOCA(nred*sizeof(IndArray)); - rhead = (int *)ALLOCA(col*sizeof(int)); + imat = (IndArray *)MALLOC(nred*sizeof(IndArray)); + rhead = (int *)MALLOC(col*sizeof(int)); for ( i = 0; i < col; i++ ) rhead[i] = 0; /* construction of index arrays */ if ( DP_Print ) { - fprintf(stderr,"%dx%d,",nsp+nred,col); + fprintf(asir_out,"%dx%d,",nsp+nred,col); } - rvect = (NM_ind_pair *)ALLOCA(nred*sizeof(NM_ind_pair)); - s0hash = (int *)ALLOCA(col*sizeof(int)); + rvect = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair)); + s0hash = (int *)MALLOC(col*sizeof(int)); for ( i = 0, s = s0vect; i < col; i++, s += nd_wpd ) s0hash[i] = ndl_hash_value(s); for ( rp = rp0, i = 0; rp; i++, rp = NEXT(rp) ) { rvect[i] = (NM_ind_pair)BDY(rp); - imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,s0hash,rvect[i]); + imat[i] = nm_ind_pair_to_vect_compress(trace,s0vect,col,s0hash,rvect[i]); rhead[imat[i]->head] = 1; } - if ( m > 0 || m == -1 ) + if ( m > 0 ) +#if defined(__GNUC__) + r0 = nd_f4_red_mod64_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); +#else r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); +#endif + else if ( m == -1 ) + r0 = nd_f4_red_sf_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred,nz); else if ( m == -2 ) - r0 = nd_f4_red_lf_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred); + r0 = nd_f4_red_lf_main(m,sp0,nsp,trace,s0vect,col,rvect,rhead,imat,nred); else +#if defined(VISUAL) + r0 = nd_f4_red_q_main(sp0,nsp,trace,s0vect,col,rvect,rhead,imat,nred); +#else r0 = nd_f4_red_gz_main(sp0,nsp,trace,s0vect,col,rvect,rhead,imat,nred); +#endif +#if 0 if ( DP_Print ) print_eg("search",&eg_search); +#endif return r0; } -/* for small finite fields */ +/* for Fp, 2<=p<2^16 */ + NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz) { @@ -7150,10 +7274,10 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s spcol = col-nred; get_eg(&eg0); /* elimination (1st step) */ - spmat = (int **)ALLOCA(nsp*sizeof(UINT *)); - svect = (UINT *)ALLOCA(col*sizeof(UINT)); - spsugar = (int *)ALLOCA(nsp*sizeof(int)); - spactive = !nz?0:(ND_pairs *)ALLOCA(nsp*sizeof(ND_pairs)); + spmat = (int **)MALLOC(nsp*sizeof(UINT *)); + svect = (UINT *)MALLOC(col*sizeof(UINT)); + spsugar = (int *)MALLOC(nsp*sizeof(int)); + spactive = !nz?0:(ND_pairs *)MALLOC(nsp*sizeof(ND_pairs)); for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { nd_sp(m,0,sp,&spol); if ( !spol ) continue; @@ -7176,14 +7300,14 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s } get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime); + fprintf(asir_out,"elim1=%.3fsec,",eg_f4_1.exectime+eg_f4_1.gctime); fflush(asir_out); } /* free index arrays */ for ( i = 0; i < nred; i++ ) GCFREE(imat[i]->index.c); /* elimination (2nd step) */ - colstat = (int *)ALLOCA(spcol*sizeof(int)); + colstat = (int *)MALLOC(spcol*sizeof(int)); if ( m == -1 ) rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat); else @@ -7201,10 +7325,10 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime); - fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", nsp,nred,sprow,spcol,rank); - fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); } if ( nz ) { for ( i = 0; i < rank-1; i++ ) NEXT(spactive[i]) = spactive[i+1]; @@ -7217,7 +7341,175 @@ NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s return r0; } -NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, +#if defined(__GNUC__) +/* for Fp, 2^15=index.c); + + /* elimination (2nd step) */ + colstat = (int *)MALLOC(spcol*sizeof(int)); + rank = nd_gauss_elim_mod64(spmat,spsugar,spactive,sprow,spcol,m,colstat); + r0 = 0; + for ( i = 0; i < rank; i++ ) { + NEXTNODE(r0,r); BDY(r) = + (pointer)vect64_to_ndv(spmat[i],spcol,col,rhead,s0vect); + SG((NDV)BDY(r)) = spsugar[i]; + GCFREE(spmat[i]); + } + if ( r0 ) NEXT(r) = 0; + + for ( ; i < sprow; i++ ) GCFREE(spmat[i]); + get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); + init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); + if ( DP_Print ) { + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + nsp,nred,sprow,spcol,rank); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); + } + if ( nz ) { + for ( i = 0; i < rank-1; i++ ) NEXT(spactive[i]) = spactive[i+1]; + if ( rank > 0 ) { + NEXT(spactive[rank-1]) = 0; + *nz = spactive[0]; + } else + *nz = 0; + } + return r0; +} +#endif + +/* for small finite fields */ + +NODE nd_f4_red_sf_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, + NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred,ND_pairs *nz) +{ + int spcol,sprow,a; + int i,j,k,l,rank; + NODE r0,r; + ND_pairs sp; + ND spol; + int **spmat; + UINT *svect,*v; + int *colstat; + struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2; + int maxrs; + int *spsugar; + ND_pairs *spactive; + + spcol = col-nred; + get_eg(&eg0); + /* elimination (1st step) */ + spmat = (int **)MALLOC(nsp*sizeof(UINT *)); + svect = (UINT *)MALLOC(col*sizeof(UINT)); + spsugar = (int *)MALLOC(nsp*sizeof(int)); + spactive = !nz?0:(ND_pairs *)MALLOC(nsp*sizeof(ND_pairs)); + for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { + nd_sp(m,0,sp,&spol); + if ( !spol ) continue; + nd_to_vect(m,s0vect,col,spol,svect); + maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rvect,nred); + for ( i = 0; i < col; i++ ) if ( svect[i] ) break; + if ( i < col ) { + spmat[sprow] = v = (UINT *)MALLOC_ATOMIC(spcol*sizeof(UINT)); + for ( j = k = 0; j < col; j++ ) + if ( !rhead[j] ) v[k++] = svect[j]; + spsugar[sprow] = MAX(maxrs,SG(spol)); + if ( nz ) + spactive[sprow] = sp; + sprow++; + } + nd_free(spol); + } + get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); + if ( DP_Print ) { + fprintf(asir_out,"elim1=%.3fsec,",eg_f4_1.exectime+eg_f4_1.gctime); + fflush(asir_out); + } + /* free index arrays */ + for ( i = 0; i < nred; i++ ) GCFREE(imat[i]->index.c); + + /* elimination (2nd step) */ + colstat = (int *)MALLOC(spcol*sizeof(int)); + rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat); + r0 = 0; + for ( i = 0; i < rank; i++ ) { + NEXTNODE(r0,r); BDY(r) = + (pointer)vect_to_ndv(spmat[i],spcol,col,rhead,s0vect); + SG((NDV)BDY(r)) = spsugar[i]; + GCFREE(spmat[i]); + } + if ( r0 ) NEXT(r) = 0; + + for ( ; i < sprow; i++ ) GCFREE(spmat[i]); + get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); + init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); + if ( DP_Print ) { + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + nsp,nred,sprow,spcol,rank); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); + } + if ( nz ) { + for ( i = 0; i < rank-1; i++ ) NEXT(spactive[i]) = spactive[i+1]; + if ( rank > 0 ) { + NEXT(spactive[rank-1]) = 0; + *nz = spactive[0]; + } else + *nz = 0; + } + return r0; +} + +NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,int trace,UINT *s0vect,int col, NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred) { int spcol,sprow,a; @@ -7236,14 +7528,14 @@ NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,UINT spcol = col-nred; get_eg(&eg0); /* elimination (1st step) */ - spmat = (mpz_t **)ALLOCA(nsp*sizeof(mpz_t *)); - svect = (mpz_t *)ALLOCA(col*sizeof(mpz_t)); - spsugar = (int *)ALLOCA(nsp*sizeof(int)); + spmat = (mpz_t **)MALLOC(nsp*sizeof(mpz_t *)); + svect = (mpz_t *)MALLOC(col*sizeof(mpz_t)); + spsugar = (int *)MALLOC(nsp*sizeof(int)); for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { - nd_sp(m,0,sp,&spol); + nd_sp(m,trace,sp,&spol); if ( !spol ) continue; nd_to_vect_lf(s0vect,col,spol,svect); - maxrs = ndv_reduce_vect_lf(svect,col,imat,rvect,nred); + maxrs = ndv_reduce_vect_lf(svect,trace,col,imat,rvect,nred); for ( i = 0; i < col; i++ ) if ( mpz_sgn(svect[i]) ) break; if ( i < col ) { spmat[sprow] = v = (mpz_t *)MALLOC(spcol*sizeof(mpz_t)); @@ -7256,16 +7548,16 @@ NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,UINT } get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime); + fprintf(asir_out,"elim1=%.3fsec,",eg_f4_1.exectime+eg_f4_1.gctime); fflush(asir_out); } /* free index arrays */ /* for ( i = 0; i < nred; i++ ) GCFREE(imat[i]->index.c); */ /* elimination (2nd step) */ - colstat = (int *)ALLOCA(spcol*sizeof(int)); + colstat = (int *)MALLOC(spcol*sizeof(int)); rank = nd_gauss_elim_lf(spmat,spsugar,sprow,spcol,colstat); - w = (pointer *)ALLOCA(rank*sizeof(pointer)); + w = (pointer *)MALLOC(rank*sizeof(pointer)); for ( i = 0; i < rank; i++ ) { #if 0 w[rank-i-1] = (pointer)vect_to_ndv_lf(spmat[i],spcol,col,rhead,s0vect); @@ -7291,10 +7583,10 @@ NODE nd_f4_red_lf_main(int m,ND_pairs sp0,int nsp,UINT get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime); - fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", nsp,nred,sprow,spcol,rank); - fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); } return r0; } @@ -7319,9 +7611,9 @@ NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,int trace,U spcol = col-nred; get_eg(&eg0); /* elimination (1st step) */ - spmat = (Q **)ALLOCA(nsp*sizeof(Q *)); - svect = (Q *)ALLOCA(col*sizeof(Q)); - spsugar = (int *)ALLOCA(nsp*sizeof(int)); + spmat = (Q **)MALLOC(nsp*sizeof(Q *)); + svect = (Q *)MALLOC(col*sizeof(Q)); + spsugar = (int *)MALLOC(nsp*sizeof(int)); for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { nd_sp(0,trace,sp,&spol); if ( !spol ) continue; @@ -7339,16 +7631,16 @@ NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,int trace,U } get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime); + fprintf(asir_out,"elim1=%.3fsec,",eg_f4_1.exectime+eg_f4_1.gctime); fflush(asir_out); } /* free index arrays */ /* for ( i = 0; i < nred; i++ ) GCFREE(imat[i]->index.c); */ /* elimination (2nd step) */ - colstat = (int *)ALLOCA(spcol*sizeof(int)); + colstat = (int *)MALLOC(spcol*sizeof(int)); rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat); - w = (pointer *)ALLOCA(rank*sizeof(pointer)); + w = (pointer *)MALLOC(rank*sizeof(pointer)); for ( i = 0; i < rank; i++ ) { #if 0 w[rank-i-1] = (pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect); @@ -7373,10 +7665,10 @@ NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,int trace,U get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime); - fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", nsp,nred,sprow,spcol,rank); - fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); } return r0; } @@ -7400,9 +7692,9 @@ NODE nd_f4_red_gz_main(ND_pairs sp0,int nsp,int trace, spcol = col-nred; get_eg(&eg0); /* elimination (1st step) */ - spmat = (GZ **)ALLOCA(nsp*sizeof(GZ *)); - svect = (GZ *)ALLOCA(col*sizeof(GZ)); - spsugar = (int *)ALLOCA(nsp*sizeof(int)); + spmat = (GZ **)MALLOC(nsp*sizeof(GZ *)); + svect = (GZ *)MALLOC(col*sizeof(GZ)); + spsugar = (int *)MALLOC(nsp*sizeof(int)); for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { nd_sp(0,trace,sp,&spol); if ( !spol ) continue; @@ -7421,16 +7713,16 @@ NODE nd_f4_red_gz_main(ND_pairs sp0,int nsp,int trace, } get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); if ( DP_Print ) { - fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime); + fprintf(asir_out,"elim1=%.3fsec,",eg_f4_1.exectime+eg_f4_1.gctime); fflush(asir_out); } /* free index arrays */ /* for ( i = 0; i < nred; i++ ) GCFREE(imat[i]->index.c); */ /* elimination (2nd step) */ - colstat = (int *)ALLOCA(spcol*sizeof(int)); + colstat = (int *)MALLOC(spcol*sizeof(int)); rank = nd_gauss_elim_gz(spmat,spsugar,sprow,spcol,colstat); - w = (pointer *)ALLOCA(rank*sizeof(pointer)); + w = (pointer *)MALLOC(rank*sizeof(pointer)); for ( i = 0; i < rank; i++ ) { #if 0 w[rank-i-1] = (pointer)vect_to_ndv_gz(spmat[i],spcol,col,rhead,s0vect); @@ -7458,10 +7750,10 @@ NODE nd_f4_red_gz_main(ND_pairs sp0,int nsp,int trace, get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); if ( DP_Print ) { - fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime); - fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", + fprintf(asir_out,"elim2=%.3fsec,",eg_f4_2.exectime+eg_f4_2.gctime); + fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", nsp,nred,sprow,spcol,rank); - fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); + fprintf(asir_out,"%.3fsec,",eg_f4.exectime+eg_f4.gctime); } return r0; } @@ -7492,8 +7784,8 @@ NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vec row = nsp+nred; /* make the matrix */ - mat = (Q **)ALLOCA(row*sizeof(Q *)); - sugar = (int *)ALLOCA(row*sizeof(int)); + mat = (Q **)MALLOC(row*sizeof(Q *)); + sugar = (int *)MALLOC(row*sizeof(int)); for ( row = a = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { nd_sp(0,0,sp,&spol); if ( !spol ) continue; @@ -7507,7 +7799,7 @@ NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vec sugar[row] = rvect[i]->sugar; } /* elimination */ - colstat = (int *)ALLOCA(col*sizeof(int)); + colstat = (int *)MALLOC(col*sizeof(int)); rank = nd_gauss_elim_q(mat,sugar,row,col,colstat); r0 = 0; for ( i = 0; i < rank; i++ ) { @@ -7599,189 +7891,6 @@ NDV nd_recv_ndv() } } -int ox_exec_f4_red(Q proc) -{ - Obj obj; - STRING fname; - NODE arg; - int s; - extern int ox_need_conv,ox_file_io; - - MKSTR(fname,"nd_exec_f4_red"); - arg = mknode(2,proc,fname); - Pox_cmo_rpc(arg,&obj); - s = get_ox_server_id(QTOS(proc)); - nd_write = iofp[s].out; - nd_read = iofp[s].in; - ox_need_conv = ox_file_io = 0; - return s; -} - -#if 0 -NODE nd_f4_red_dist(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0,ND_pairs *nz) -{ - int nsp,nred; - int i,rank,s; - NODE rp,r0,r; - ND_pairs sp; - NM_ind_pair pair; - NMV nmv; - NM nm; - NDV nf; - Obj proc,dmy; - - ox_launch_main(0,0,&proc); - s = ox_exec_f4_red((Q)proc); - - nd_send_int(m); - nd_send_int(nd_nvar); - nd_send_int(nd_bpe); - nd_send_int(nd_wpd); - nd_send_int(nmv_adv); - - saveobj(nd_write,dp_current_spec->obj); fflush(nd_write); - - nd_send_int(nd_psn); - for ( i = 0; i < nd_psn; i++ ) nd_send_ndv(nd_ps[i]); - - for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); - nd_send_int(nsp); - for ( i = 0, sp = sp0; i < nsp; i++, sp = NEXT(sp) ) { - nd_send_int(sp->i1); nd_send_int(sp->i2); - } - - nd_send_int(col); nd_send_intarray(s0vect,col*nd_wpd); - - nred = length(rp0); nd_send_int(nred); - for ( i = 0, rp = rp0; i < nred; i++, rp = NEXT(rp) ) { - pair = (NM_ind_pair)BDY(rp); - nd_send_int(pair->index); - nd_send_intarray(pair->mul->dl,nd_wpd); - } - fflush(nd_write); - rank = nd_recv_int(); - fprintf(asir_out,"rank=%d\n",rank); - r0 = 0; - for ( i = 0; i < rank; i++ ) { - nf = nd_recv_ndv(); - NEXTNODE(r0,r); BDY(r) = (pointer)nf; - } - Pox_shutdown(mknode(1,proc),&dmy); - return r0; -} - -/* server side */ - -void nd_exec_f4_red_dist() -{ - int m,i,nsp,col,s0size,nred,spcol,j,k; - NM_ind_pair *rp0; - NDV nf; - UINT *s0vect; - IndArray *imat; - int *rhead; - int **spmat; - UINT *svect,*v; - ND_pairs *sp0; - int *colstat; - int a,sprow,rank; - struct order_spec *ord; - Obj ordspec; - ND spol; - int maxrs; - int *spsugar; - - nd_read = iofp[0].in; - nd_write = iofp[0].out; - m = nd_recv_int(); - nd_nvar = nd_recv_int(); - nd_bpe = nd_recv_int(); - nd_wpd = nd_recv_int(); - nmv_adv = nd_recv_int(); - - loadobj(nd_read,&ordspec); - create_order_spec(0,ordspec,&ord); - nd_init_ord(ord); - nd_setup_parameters(nd_nvar,0); - - nd_psn = nd_recv_int(); - nd_ps = (NDV *)MALLOC(nd_psn*sizeof(NDV)); - nd_bound = (UINT **)MALLOC(nd_psn*sizeof(UINT *)); - for ( i = 0; i < nd_psn; i++ ) { - nd_ps[i] = nd_recv_ndv(); - nd_bound[i] = ndv_compute_bound(nd_ps[i]); - } - - nsp = nd_recv_int(); - sp0 = (ND_pairs *)MALLOC(nsp*sizeof(ND_pairs)); - for ( i = 0; i < nsp; i++ ) { - NEWND_pairs(sp0[i]); - sp0[i]->i1 = nd_recv_int(); sp0[i]->i2 = nd_recv_int(); - ndl_lcm(HDL(nd_ps[sp0[i]->i1]),HDL(nd_ps[sp0[i]->i2]),LCM(sp0[i])); - } - - col = nd_recv_int(); - s0size = col*nd_wpd; - s0vect = (UINT *)MALLOC(s0size*sizeof(UINT)); - nd_recv_intarray(s0vect,s0size); - - nred = nd_recv_int(); - rp0 = (NM_ind_pair *)MALLOC(nred*sizeof(NM_ind_pair)); - for ( i = 0; i < nred; i++ ) { - rp0[i] = (NM_ind_pair)MALLOC(sizeof(struct oNM_ind_pair)); - rp0[i]->index = nd_recv_int(); - rp0[i]->mul = (NM)MALLOC(sizeof(struct oNM)+(nd_wpd-1)*sizeof(UINT)); - nd_recv_intarray(rp0[i]->mul->dl,nd_wpd); - } - - spcol = col-nred; - imat = (IndArray *)MALLOC(nred*sizeof(IndArray)); - rhead = (int *)MALLOC(col*sizeof(int)); - for ( i = 0; i < col; i++ ) rhead[i] = 0; - - /* construction of index arrays */ - for ( i = 0; i < nred; i++ ) { - imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rp0[i]); - rhead[imat[i]->head] = 1; - } - - /* elimination (1st step) */ - spmat = (int **)MALLOC(nsp*sizeof(UINT *)); - svect = (UINT *)MALLOC(col*sizeof(UINT)); - spsugar = (int *)ALLOCA(nsp*sizeof(int)); - for ( a = sprow = 0; a < nsp; a++ ) { - nd_sp(m,0,sp0[a],&spol); - if ( !spol ) continue; - nd_to_vect(m,s0vect,col,spol,svect); - if ( m == -1 ) - maxrs = ndv_reduce_vect_sf(m,svect,col,imat,rp0,nred); - else - maxrs = ndv_reduce_vect(m,svect,col,imat,rp0,nred); - for ( i = 0; i < col; i++ ) if ( svect[i] ) break; - if ( i < col ) { - spmat[sprow] = v = (UINT *)MALLOC(spcol*sizeof(UINT)); - for ( j = k = 0; j < col; j++ ) - if ( !rhead[j] ) v[k++] = svect[j]; - spsugar[sprow] = MAX(maxrs,SG(spol)); - sprow++; - } - nd_free(spol); - } - /* elimination (2nd step) */ - colstat = (int *)ALLOCA(spcol*sizeof(int)); - if ( m == -1 ) - rank = nd_gauss_elim_sf(spmat,spsugar,sprow,spcol,m,colstat); - else - rank = nd_gauss_elim_mod(spmat,spsugar,0,sprow,spcol,m,colstat); - nd_send_int(rank); - for ( i = 0; i < rank; i++ ) { - nf = vect_to_ndv(spmat[i],spcol,col,rhead,s0vect); - nd_send_ndv(nf); - } - fflush(nd_write); -} -#endif - int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat) { int i,j,t,c,rank,inv; @@ -7894,6 +8003,85 @@ int nd_gauss_elim_mod(int **mat0,int *sugar,ND_pairs * return rank; } +#if defined(__GNUC__) + +int nd_gauss_elim_mod64(U64 **mat,int *sugar,ND_pairs *spactive,int row,int col,int md,int *colstat) +{ + int i,j,k,l,rank,s; + U64 inv; + U64 a; + UINT c; + U64 *t,*pivot,*pk; + UINT *ck; + UINT **cmat; + UINT *ct; + ND_pairs pair; + + cmat = (UINT **)MALLOC(row*sizeof(UINT *)); + for ( i = 0; i < row; i++ ) { + cmat[i] = MALLOC_ATOMIC(col*sizeof(UINT)); + bzero(cmat[i],col*sizeof(UINT)); + } + + for ( rank = 0, j = 0; j < col; j++ ) { + for ( i = rank; i < row; i++ ) { + a = mat[i][j]; c = cmat[i][j]; + MOD128(a,c,md); + mat[i][j] = a; cmat[i][j] = 0; + } + for ( i = rank; i < row; i++ ) + if ( mat[i][j] ) + break; + if ( i == row ) { + colstat[j] = 0; + continue; + } else + colstat[j] = 1; + if ( i != rank ) { + t = mat[i]; mat[i] = mat[rank]; mat[rank] = t; + ct = cmat[i]; cmat[i] = cmat[rank]; cmat[rank] = ct; + s = sugar[i]; sugar[i] = sugar[rank]; sugar[rank] = s; + if ( spactive ) { + pair = spactive[i]; spactive[i] = spactive[rank]; + spactive[rank] = pair; + } + } + /* column j is normalized */ + s = sugar[rank]; + inv = invm((UINT)mat[rank][j],md); + /* normalize pivot row */ + for ( k = j, pk = mat[rank]+j, ck = cmat[rank]+j; k < col; k++, pk++, ck++ ) { + a = *pk; c = *ck; MOD128(a,c,md); *pk = (a*inv)%md; *ck = 0; + } + for ( i = rank+1; i < row; i++ ) { + if ( a = mat[i][j] ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect64(md,mat[i]+j,cmat[i]+j,mat[rank]+j,(int)(md-a),col-j); + } + } + rank++; + } + for ( j = col-1, l = rank-1; j >= 0; j-- ) + if ( colstat[j] ) { + for ( k = j, pk = mat[l]+j, ck = cmat[l]+j; k < col; k++, pk++, ck++ ) { + a = *pk; c = *ck; MOD128(a,c,md); *pk = a; *ck = 0; + } + s = sugar[l]; + for ( i = 0; i < l; i++ ) { + a = mat[i][j]; c = cmat[i][j]; MOD128(a,c,md); mat[i][j] = a; cmat[i][j] = 0; + if ( a ) { + sugar[i] = MAX(sugar[i],s); + red_by_vect64(md,mat[i]+j,cmat[i]+j,mat[l]+j,(int)(md-a),col-j); + } + } + l--; + } + for ( i = 0; i < row; i++ ) GCFREE(cmat[i]); + GCFREE(cmat); + return rank; +} +#endif + int nd_gauss_elim_sf(int **mat0,int *sugar,int row,int col,int md,int *colstat) { int i,j,k,l,inv,a,rank,s; @@ -8183,7 +8371,7 @@ void nd_det(int mod,MAT f,P *rp) chsgnq(ONE,&mone); for ( j = 0, sgn = 1; j < n; j++ ) { if ( DP_Print ) { - fprintf(stderr,".",j); + fprintf(asir_out,".",j); } for ( i = j; i < n && !dm[i][j]; i++ ); if ( i == n ) { @@ -8213,14 +8401,14 @@ void nd_det(int mod,MAT f,P *rp) nd_det_reconstruct(dm,n,j,d); for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) { -/* if ( DP_Print ) fprintf(stderr," i=%d\n ",i); */ +/* if ( DP_Print ) fprintf(asir_out," i=%d\n ",i); */ mi = dm[i]; mij = mi[j]; if ( mod ) ndv_mul_c(mod,mij,mod-1); else ndv_mul_c_q(mij,mone); for ( k = j+1; k < n; k++ ) { -/* if ( DP_Print ) fprintf(stderr,"k=%d ",k); */ +/* if ( DP_Print ) fprintf(asir_out,"k=%d ",k); */ bucket = create_pbucket(); if ( mi[k] ) { nmv = BDY(mjj); len = LEN(mjj); @@ -8239,12 +8427,12 @@ void nd_det(int mod,MAT f,P *rp) u = nd_quo(mod,bucket,d); mi[k] = ndtondv(mod,u); } -/* if ( DP_Print ) fprintf(stderr,"\n",k); */ +/* if ( DP_Print ) fprintf(asir_out,"\n",k); */ } d = mjj; } if ( DP_Print ) { - fprintf(stderr,"\n",k); + fprintf(asir_out,"\n",k); } if ( sgn < 0 ) if ( mod ) @@ -8420,14 +8608,14 @@ int nd_monic(int mod,ND *p) is_lc = 1; while ( 1 ) { NEWMP(mp0); mp = mp0; - mp->c = (P)CQ(m); + mp->c = (Obj)CQ(m); mp->dl = nd_separate_d(DL(m),DL(ma)); NEWNM(mb); for ( m = NEXT(m); m; m = NEXT(m) ) { alg = nd_separate_d(DL(m),DL(mb)); if ( !ndl_equal(DL(ma),DL(mb)) ) break; - NEXTMP(mp0,mp); mp->c = (P)CQ(m); mp->dl = alg; + NEXTMP(mp0,mp); mp->c = (Obj)CQ(m); mp->dl = alg; } NEXT(mp) = 0; MKDP(nd_nalg,mp0,nm); @@ -8575,7 +8763,8 @@ void parse_nd_option(NODE opt) nd_newelim = 0; nd_intersect = 0; nd_nzlist = 0; nd_splist = 0; nd_check_splist = 0; nd_sugarweight = 0; - + nd_f4red =0; + nd_rank0 = 0; for ( t = opt; t; t = NEXT(t) ) { p = BDY((LIST)BDY(t)); key = BDY((STRING)BDY(p)); @@ -8587,27 +8776,34 @@ void parse_nd_option(NODE opt) else if ( !strcmp(key,"nora") ) nd_nora = value?1:0; else if ( !strcmp(key,"gbblock") ) { - if ( !value || OID(value) != O_LIST ) - error("nd_* : invalid value for gbblock option"); - u = BDY((LIST)value); - nd_gbblock = MALLOC((2*length(u)+1)*sizeof(int)); - for ( i = 0; u; u = NEXT(u) ) { - p = BDY((LIST)BDY(u)); - s = nd_gbblock[i++] = QTOS((Q)BDY(p)); - nd_gbblock[i++] = s+QTOS((Q)BDY(NEXT(p)))-1; - } - nd_gbblock[i] = -1; + if ( value && OID(value) == O_LIST ) { + u = BDY((LIST)value); + nd_gbblock = MALLOC((2*length(u)+1)*sizeof(int)); + for ( i = 0; u; u = NEXT(u) ) { + p = BDY((LIST)BDY(u)); + s = nd_gbblock[i++] = QTOS((Q)BDY(p)); + nd_gbblock[i++] = s+QTOS((Q)BDY(NEXT(p)))-1; + } + nd_gbblock[i] = -1; + } else + nd_gbblock = 0; } else if ( !strcmp(key,"newelim") ) nd_newelim = value?1:0; else if ( !strcmp(key,"intersect") ) nd_intersect = value?1:0; + else if ( !strcmp(key,"lf") ) + nd_lf = value?1:0; else if ( !strcmp(key,"trace") ) { u = BDY((LIST)value); nd_nzlist = BDY((LIST)ARG2(u)); nd_bpe = QTOS((Q)ARG3(u)); - } else if ( !strcmp(key,"splist") ) + } else if ( !strcmp(key,"f4red") ) { + nd_f4red = QTOS((Q)value); + } else if ( !strcmp(key,"rank0") ) { + nd_rank0 = value?1:0; + } else if ( !strcmp(key,"splist") ) { nd_splist = value?1:0; - else if ( !strcmp(key,"check_splist") ) { + } else if ( !strcmp(key,"check_splist") ) { nd_check_splist = BDY((LIST)value); } else if ( !strcmp(key,"sugarweight") ) { u = BDY((LIST)value); @@ -8798,15 +8994,11 @@ MAT nd_btog(LIST f,LIST v,int mod,struct order_spec *o printf("%d ",i); fflush(stdout); ti = BDY((LIST)BDY(t)); p[j=QTOS((Q)ARG0(ti))] = btog(BDY((LIST)ARG1(ti)),p,nb,mod); - if ( j == 441 ) - printf("afo"); } for ( t = intred, i=0; t; t = NEXT(t), i++ ) { printf("%d ",i); fflush(stdout); ti = BDY((LIST)BDY(t)); p[j=QTOS((Q)ARG0(ti))] = btog(BDY((LIST)ARG1(ti)),p,nb,mod); - if ( j == 441 ) - printf("afo"); } m = length(ind); MKMAT(mat,nb,m); @@ -8915,3 +9107,228 @@ void ndv_print_lf(NDV p) printf("\n"); } } + +void nd_f4_lf_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp) +{ + VL tv,fv,vv,vc,av; + NODE fd,fd0,in0,in,r,r0,t,s,cand,alist; + int m,nocheck,nvar,mindex,e,max; + NDV c; + NMV a; + P p,zp; + Q dmy; + EPOS oepos; + int obpe,oadv,wmax,i,len,cbpe,ishomo,nalg,mrank,trank,ompos; + Alg alpha,dp; + P poly; + LIST f1,f2,zpl; + Obj obj; + NumberField nf; + struct order_spec *ord1; + struct oEGT eg_check,eg0,eg1; + NODE tr,tl1,tl2,tl3,tl4; + LIST l1,l2,l3,l4,l5; + int *perm; + int j,ret; + Q jq,bpe; + + nd_module = 0; + parse_nd_option(current_option); + get_vars((Obj)f,&fv); pltovl(v,&vv); vlminus(fv,vv,&nd_vc); + if ( nd_vc ) + error("nd_f4_lf_trace : computation over a rational function field is not implemented"); + for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ ); + switch ( ord->id ) { + case 1: + if ( ord->nv != nvar ) + error("nd_f4_lf_trace : invalid order specification"); + break; + default: + break; + } + + nd_ntrans = nvar; + nd_nalg = 0; + + nocheck = 0; + mindex = 0; + + /* do not use on-demand load/save */ + nd_demand = 0; + m = trace > 1 ? trace : get_lprime(mindex); + nd_init_ord(ord); + mrank = 0; + for ( t = BDY(f), max = 1; t; t = NEXT(t) ) + for ( tv = vv; tv; tv = NEXT(tv) ) { + if ( nd_module ) { + s = BDY((LIST)BDY(t)); + trank = length(s); + mrank = MAX(mrank,trank); + for ( ; s; s = NEXT(s) ) { + e = getdeg(tv->v,(P)BDY(s)); + max = MAX(e,max); + } + } else { + e = getdeg(tv->v,(P)BDY(t)); + max = MAX(e,max); + } + } + nd_setup_parameters(nvar,max); + obpe = nd_bpe; oadv = nmv_adv; oepos = nd_epos; ompos = nd_mpos; + ishomo = 1; + /* XXX */ + for ( in0 = 0, fd0 = 0, t = BDY(f); t; t = NEXT(t) ) { + if ( nd_module ) { + c = (pointer)pltondv(CO,vv,(LIST)BDY(t)); + } else { + c = (pointer)ptondv(CO,vv,(P)BDY(t)); + } + if ( ishomo ) + ishomo = ishomo && ndv_ishomo(c); + if ( c ) { + NEXTNODE(in0,in); BDY(in) = (pointer)c; + NEXTNODE(fd0,fd); BDY(fd) = (pointer)ndv_dup(0,c); + } + } + if ( in0 ) NEXT(in) = 0; + if ( fd0 ) NEXT(fd) = 0; + if ( !ishomo && homo ) { + for ( t = in0, wmax = max; t; t = NEXT(t) ) { + c = (NDV)BDY(t); len = LEN(c); + for ( a = BDY(c), i = 0; i < len; i++, NMV_ADV(a) ) + wmax = MAX(TD(DL(a)),wmax); + } + homogenize_order(ord,nvar,&ord1); + nd_init_ord(ord1); + nd_setup_parameters(nvar+1,wmax); + for ( t = fd0; t; t = NEXT(t) ) + ndv_homogenize((NDV)BDY(t),obpe,oadv,oepos,ompos); + } + if ( MaxDeg > 0 ) nocheck = 1; + ret = ndv_setup(-2,m,fd0,nd_gbblock?1:0,0); + if ( ret ) + cand = nd_f4_lf_trace_main(m,&perm); + if ( !ret || !cand ) { + *rp = 0; return; + } + if ( !ishomo && homo ) { + /* dehomogenization */ + for ( t = cand; t; t = NEXT(t) ) ndv_dehomogenize((NDV)BDY(t),ord); + nd_init_ord(ord); + nd_setup_parameters(nvar,0); + } + cand = ndv_reducebase(cand,perm); + cand = ndv_reduceall(-2,cand); + cbpe = nd_bpe; + get_eg(&eg0); + if ( ret = ndv_check_membership(-2,in0,obpe,oadv,oepos,cand) ) { + /* gbcheck : cand is a GB of Id(cand) ? */ + ret = nd_f4(-2,0,0); + } + if ( !ret ) { + /* failure */ + *rp = 0; return; + } + get_eg(&eg1); init_eg(&eg_check); add_eg(&eg_check,&eg0,&eg1); + if ( DP_Print ) + fprintf(asir_out,"check=%.3fsec\n",eg_check.exectime+eg_check.gctime); + /* dp->p */ + nd_bpe = cbpe; + nd_setup_parameters(nd_nvar,0); + for ( r = cand; r; r = NEXT(r) ) { + if ( nd_module ) BDY(r) = ndvtopl(-2,CO,vv,BDY(r),mrank); + else BDY(r) = (pointer)ndvtop(-2,CO,vv,BDY(r)); + } + MKLIST(*rp,cand); +} + +NODE nd_f4_lf_trace_main(int m,int **indp) +{ + int i,nh,stat,index; + NODE r,rm,g; + ND_pairs d,l,l0,t; + ND spol,red; + NDV nf,redv,nfqv,nfv; + NM s0,s; + NODE rp0,srp0,nflist,nflist_lf; + int nsp,nred,col,rank,len,k,j,a; + UINT c; + UINT **spmat; + UINT *s0vect,*svect,*p,*v; + int *colstat; + IndArray *imat; + int *rhead; + int spcol,sprow; + int sugar; + PGeoBucket bucket; + struct oEGT eg0,eg1,eg_f4; + + g = 0; d = 0; + for ( i = 0; i < nd_psn; i++ ) { + d = update_pairs(d,g,i,0); + g = update_base(g,i); + } + while ( d ) { + get_eg(&eg0); + l = nd_minsugarp(d,&d); + sugar = SG(l); + if ( MaxDeg > 0 && sugar > MaxDeg ) break; + bucket = create_pbucket(); + stat = nd_sp_f4(m,0,l,bucket); + if ( !stat ) { + for ( t = l; NEXT(t); t = NEXT(t) ); + NEXT(t) = d; d = l; + d = nd_reconstruct(1,d); + continue; + } + if ( bucket->m < 0 ) continue; + col = nd_symbolic_preproc(bucket,0,&s0vect,&rp0); + if ( !col ) { + for ( t = l; NEXT(t); t = NEXT(t) ); + NEXT(t) = d; d = l; + d = nd_reconstruct(1,d); + continue; + } + get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); + if ( DP_Print ) + fprintf(asir_out,"sugar=%d,symb=%.3fsec,", + sugar,eg_f4.exectime+eg_f4.gctime); + nflist = nd_f4_red(m,l,0,s0vect,col,rp0,&l0); + if ( !l0 ) continue; + l = l0; + + /* over LF */ + bucket = create_pbucket(); + stat = nd_sp_f4(-2,1,l,bucket); + if ( !stat ) { + for ( t = l; NEXT(t); t = NEXT(t) ); + NEXT(t) = d; d = l; + d = nd_reconstruct(1,d); + continue; + } + if ( bucket->m < 0 ) continue; + col = nd_symbolic_preproc(bucket,1,&s0vect,&rp0); + if ( !col ) { + for ( t = l; NEXT(t); t = NEXT(t) ); + NEXT(t) = d; d = l; + d = nd_reconstruct(1,d); + continue; + } + nflist_lf = nd_f4_red(-2,l,1,s0vect,col,rp0,0); + /* adding new bases */ + for ( rm = nflist, r = nflist_lf; r && rm; rm = NEXT(rm), r = NEXT(r) ) { + nfv = (NDV)BDY(rm); + nfqv = (NDV)BDY(r); + if ( DL_COMPARE(HDL(nfv),HDL(nfqv)) ) return 0; + ndv_removecont(m,nfv); + ndv_removecont(-2,nfqv); + nh = ndv_newps(-2,nfv,nfqv,1); + d = update_pairs(d,g,nh,0); + g = update_base(g,nh); + } + if ( r || rm ) return 0; + } + conv_ilist(nd_demand,1,g,indp); + return g; +} +